A next task for Sting

[lech]

As I earlier wrote Sting needs still a bit of code to prove that chess engines can solve all chess postions.
It is a very hard one, sent by Nicolas.
It is a prototype, but at this moment my new idea needs less than 15 minutes of my lazy old dual core..
I think it can be the last branch which needs a special code. All the rest Stockfish should solve by high depths.
I am waitng for next tough positions (a horizon efect), but I doubt that they are yet.
When the new version of Sting ? I need a few months to fine-tune it.
[d]q1N5/b1p5/1pP2k2/pP3Bp1/P1P5/6P1/1K3P2/8 w - - bm Bd7 or Bg4 to win

Code: Select all

info depth 36 multipv 1 score cp 0 nodes 249899783 nps 973615 time 256672 pv g3g4 a8b8 b2c3 f6e5 c3d3 e5f4 d3e2 b8a8 e2d2 f4e5 d2d3 a8b8 d3e3 e5f6 e3e4 f6f7 e4e5 f7g7 f5e6 g7g6 e6f5 g6f7 f5e6 f7g6
info depth 37
info currmove g3g4 currmovenumber 1
info nodes 359716548 nps 975171 time 368875
info currmove c8a7 currmovenumber 2
info currmove b2b3 currmovenumber 3
info currmove b2c1 currmovenumber 4
info nodes 360675381 nps 974961 time 369938
info currmove b2c2 currmovenumber 5
info currmove b2b1 currmovenumber 6
info currmove f5d3 currmovenumber 7
info currmove f5c2 currmovenumber 8
info currmove b2c3 currmovenumber 9
info nodes 361075331 nps 974767 time 370422
info currmove f5e4 currmovenumber 10
info currmove f5g4 currmovenumber 11
info currmove b2a2 currmovenumber 12
info nodes 361848068 nps 974223 time 371422
info currmove f5b1 currmovenumber 13
info currmove f5d7 currmovenumber 14
info depth 37 multipv 1 score cp 8 lowerbound nodes 785370377 nps 936008 time 839063 pv f5d7
info nodes 785370377 nps 936008 time 839063
info currmove f5d7 currmovenumber 1
info depth 37 multipv 1 score cp 16 lowerbound nodes 810663207 nps 935762 time 866313 pv f5d7
info nodes 810663207 nps 935762 time 866313
info currmove f5d7 currmovenumber 1
info depth 37 multipv 1 score cp 28 lowerbound nodes 820376078 nps 935150 time 877266 pv f5d7
info nodes 820376078 nps 935133 time 877282
info currmove f5d7 currmovenumber 1
info depth 37 multipv 1 score cp 46 lowerbound nodes 825664657 nps 935083 time 882985 pv f5d7
info nodes 825664657 nps 935067 time 883000
info currmove f5d7 currmovenumber 1
info depth 37 multipv 1 score cp 73 lowerbound nodes 832994463 nps 934750 time 891141 pv f5d7
info nodes 832994463 nps 934750 time 891141
info currmove f5d7 currmovenumber 1
info depth 37 multipv 1 score cp 114 lowerbound nodes 841636083 nps 934728 time 900407 pv f5d7
info nodes 841636083 nps 934728 time 900407
info currmove f5d7 currmovenumber 1
info depth 37 multipv 1 score cp 175 lowerbound nodes 867767001 nps 930207 time 932875 pv f5d7
info nodes 867767001 nps 930207 time 932875
info currmove f5d7 currmovenumber 1
info nodes 1054721250 nps 924889 time 1140375
info currmove g3g4 currmovenumber 2
info currmove c8a7 currmovenumber 3
info currmove f5e4 currmovenumber 4
info currmove f2f4 currmovenumber 5
info currmove b2c3 currmovenumber 6
info currmove f2f3 currmovenumber 7
info currmove f5g4 currmovenumber 8
info currmove f5d3 currmovenumber 9
info currmove f5h3 currmovenumber 10
info currmove b2b3 currmovenumber 11
info currmove f5e6 currmovenumber 12
info currmove b2c2 currmovenumber 13
info currmove c4c5 currmovenumber 14
info nodes 1055344661 nps 924840 time 1141110
info currmove c8e7 currmovenumber 15
info currmove f5c2 currmovenumber 16
info currmove c8d6 currmovenumber 17
info currmove b2a2 currmovenumber 18
info currmove b2c1 currmovenumber 19
info currmove b2b1 currmovenumber 20
info currmove b2a3 currmovenumber 21
info currmove f5g6 currmovenumber 22
info currmove f5b1 currmovenumber 23
info currmove b2a1 currmovenumber 24
info currmove f5h7 currmovenumber 25
info currmove c8b6 currmovenumber 26
info depth 37 multipv 1 score cp 193 nodes 1055467236 nps 924808 time 1141282 pv f5d7 f6e5 f2f3 e5f6 d7h3 f6f7 b2c3 a7b8 h3f5 f7f6 f5d7 f6g7 c3d3 b8a7 d3e4 g7f6 f3f4 a7b8 f4f5 b8a7 e4e3 f6g7 e3f3 a8b8 f3e4 g7f6 e4f3 f6g7

[lech]

Let me add a few words more.

To the branches solved by my project Sting (all aplies to a horizon effect):
A. Sacrifices leading to a long play (long and deep under beta)
B. Zugzwangs with beta = VALUE_DRAW (zero)
C. Zugzwangs delayed by neutral pawn moves
D. Trapped pieces
E. Sacrifices leading to fortress

I add now a next new one:
F. Walking King

All the branches (questions, problems) lead to the horizon effect and make solving of some chess positions, even for such a fast and deep (depths) engine like Stockfish 6, practicaly impossible. It means, there is needed a bit of code to help do it.
Is the above list of problem complete? Rather yes, and my work is very close to the finish.
Remains only one question. Sting should be a bit faster and a few works at the static evoluation is needed.

[lech]

To end this topic, I will try to explain the horizon effect and the walking King problem.
This example is based on the previuos one, but is much easier to read.
[d]1qN5/b1pB4/1pP2k2/pP3P2/P1P5/8/8/K7 w - - win
To win White has do a long way: e.g.
Ka1->b2->c3->d3->e3->f3->g4->h5->h6->h7->g8->f8->e8->d8.
Each move can be reduced directly and is reduced by the null move reduction too.
It gives a huge reduction (horizon effect) and probably no one engine in default setting can do it.
Note that Black is not able to guard this line, and each of neutral moves Q or B get beta. Black King also is sticking to the passed Pawn f5 and can't stop the white King.
This solution is rather easy for people but for engines it seems to be impossible to do without a special code.

BTW. A next Sting 6 will be probably first 32-bit demo version (time per move and infinity only) and will be contain a code for walking Kings, closed pawn positions and maybe a try to reduce the score for draw positions.

[yanquis1972]

hey marek,

just wanted to note i really appreciate your efforts w/ sting; i think as far as the public is concerned niche engines is a space that needs filling as we've been in an era of extremely strong one-engine dominance for a very long time now (tho thank god we have komodo these days , but even its advantages arent with what your engine does) , a lot of productive potential there.